-
Notifications
You must be signed in to change notification settings - Fork 1
Nokia
Object recognition algorithm to be developed for a specific type of yet unknown object.
Trained object recognition algorithms are commonly available with the predefined list of recognized objects. Providing simple/efficient means to extend the list of recognized objects with limited training data is the challenge.
You may find an example for the extension task here:
https://medium.com/practical-deep-learning/a-complete-transfer-learning-toolchain-for-semantic-segmentation-3892d722b604
You’ll be able to take photos about the new object to be recognized by the algorithm at the Hackathon venue. Generating good training data is also part of the challenge.
Install Google Tensorflow (https://www.tensorflow.org/) on your laptop with a model trained on the COCO dataset (http://cocodataset.org/), this is an example for that.
Linux Python Tensorflow
A Tweet is going to spread in the follower network in time having several re-tweeting on the same message while it reaches maximum number of followers. Some echo may help to reach more followers, some may not, so the main impacted population is changing based on the source tweet. To understand better the information spread, a type of tweet should be analyzed and understand main impacted population and maximum tweet activity in a time interval.
The quantification of the impact of a Tweet, e.g., reach (re-tweets/time window, number of recipients)
Internet connection, Twitter API (https://developer.twitter.com/en/use-cases/analyze)
javascript web api analytics
Smart public transport will have an important role in future smart cities. Identifying travel routes, the way of transports used by travelers is crucial to support route planning and public transport optimization use cases. Collecting crowd travel data (data collected by individual commuters with their mobile devices) and analyzing them provides a chance to determine the travelling behaviors of the masses. Nokia has developed an Android client that collects travel related information from friendly users’ mobile devices (e.g. location related KPIs, motion sensor data) that is uploaded and stored in a backend server. These device logs together with the Budapest public transport information (available through open APIs) will be used for the challenge.
- Find out the way of transport (which bus/tram lines used…) from the travelers’ logs. Map the offline, user collected mobile device sensor logs with BKK travel information gathered from BKK API. Create an analytical model for transport type prediction, i.e. to infer transportation type bus/tram for the sensor (e.g. acceleration) data
- Find out bus/tram stops from the travelers’ logs. Create an analytical model for predicting bus/tram stop locations.
wifi internet connection, offline data set for motion sensor logs, description of the Budapest public transport APIs
- wifi internet connection
- offline data set for motion sensor logs and location data; the format of the data:
Sendor data:
SENSOR_TIMESTAMP : unix epoch time [ms]
SENSOR_TYPE : as per Android Sensor API
SENSOR_DATA_X : sensor X value
SENSOR_DATA_Y : sensor Y value
SENSOR_DATA_Z : sensor Z value
Location data:
LOCATION_TIMESTAMP : unix epoch time [ms]
LOCATION_DATA_LAT : latitude [deg]
LOCATION_DATA_LON : longitude [deg]
LOCATION_DATA_ALT : altitude [m]
LOCATION_DATA_ACC : accuracy [m]
LOCATION_DATA_VEL : velocity [m/s]
- description of the Budapest public transport
Option 1: can be downloaded in the Google developed GTFS (General Transit Feed Specification) format from here:
http://www.bkk.hu/gtfs/budapest_gtfs.zip
Option 2: can be queried via the API of the BKK Futár (http://futar.bkk.hu/), here is the API specification:
https://bkkfutar.docs.apiary.io/#reference/0/search/search?console=1
web API usage, data mining, analytics skills, “R” or python programming skills